NVIDIA 450 Linux Beta Driver Quietly Rolls Out With New PRIME Option, Other Improvements

[phoronix]

Phoronix: NVIDIA 450 Linux Beta Driver Quietly Rolls Out With New PRIME Option, Other Improvements

NVIDIA has yet to formally announce the 450 Linux driver series in beta or stable form, but the first pre-release builds in the 450 branch did manage to creep out this past week alongside the CUDA 11.0 release candidate...

http://www.phoronix.com/scan.php?page=news_item&px=NVIDIA-450-Linux-Beta

[shmerl]

Nvidia catching up on PRIME support, decades late. The benefits of the blob.

[spykes]

Hopefully this driver series will also add the Vulkan bits needed for Valve to support async reprojection 🙏

[pipe13]

Nvidia catching up on PRIME support, decades late. The benefits of the blob.

...I see. And lack of this particular "AllowPRIMEDisplayOffloadSink" option has stopped your particular show exactly how?

I ask only because Nvidia has supported PRIME Display offloading in one form or another at least since 435.17, and am curious as to why lack of this particular option has rendered your dearly purchased Nvidia hardware so utterly unsuited to your task.

It seems to me Display Offload is a somewhat niche requirement for portable devices that simply *must* render a few gpu-intensive tasks while on battery power. So it's certainly a "nice to have." But no corporation has unlimited manpower and resources, and their release notes suggest Nvidia has few idle minds and hands.

https://download.nvidia.com/XFree86/...E/randr14.html
https://download.nvidia.com/XFree86/...eroffload.html

[mdedetrich]

Nvidia catching up on PRIME support, decades late. The benefits of the blob.

The same blob that gave Windows PRIME support for decades?

This has less to do with the fact that its a blob and more to do with the fact we are dealing with Linux and X11/Wayland

[lateo]

Nvidia catching up on PRIME support, decades late. The benefits of the blob.

Nvidia was (and probably still is) just not interested in supporting "non-corporate" linux computing.

Is AMD really making it's way in the new computers and thus somehow "threatening" Nvidia for the latter to finally make a move ?
Sounds nice but unrealistic

Edit: This may have something to do with moves from guys like Lenovo (planning to sell some hardware preinstalled with ubuntu, red hat EL and a third i can't remember [edit: the third was windows] : it seems you would select RHEL just as easily as your are now selecting win home rather than win pro). Suddenly, hardware has to be supported on linux to be integrated and sold.
Press release (2020/06/02) : https://news.lenovo.com/pressroom/pr...ta-scientists/

[shmerl]

...I see. And lack of this particular "AllowPRIMEDisplayOffloadSink" option has stopped your particular show exactly how?

Should we care how? Nvidia couldn't implement PRIME for ages, since they refused to upstream their driver. Their reasons is not something I really care about. A blob is a blob and lack of integration with upstream is totally expected from it.

[abott]

I've not tried to run Linux on it as Nvidia was always terrible when I demo'd it. Does their offloading work with DRI now? Is it good enough to actually make my laptop usable with Linux? I'd love to move off of Win10 on my laptop, but Nvidia power consumption was always the blocker.

[mdedetrich]

I've not tried to run Linux on it as Nvidia was always terrible when I demo'd it. Does their offloading work with DRI now? Is it good enough to actually make my laptop usable with Linux? I'd love to move off of Win10 on my laptop, but Nvidia power consumption was always the blocker.

I have been using NVdia Prime Offload and it does work as advertised, there are some caveats though

1. NVidia Prime Offload cannot turn off the GPU completely like it can in Windows however this is apparently due to a Linux kernel limitation. Note that even though Prime Render Offload doesn't completely turn off the GPU it still uses a lot less power when its not being used (~5 watts) so its not pointless (it achieves its goal of using a lot less power compared to having both GPU and CPU constantly on).
2. You have to explicitly specify what applications you want the GPU to use by setting an environment variable as you launch that application. In other words its manual unlike Windows (although Bumblebee was the exact same). I think this may be more of a Linux problem than an NVidia problem because afaik there is no way in Linux to figure out "ss this application demanding for a GPU". Might make sense for freedesktop to add a flag for `.desktop` files which specify if this application is "GPU demanding" and handle it that way?

Also GPU video acceleration (VDPAU) works fine, again you just need to enable prime render offload when running applications like chrome/firefox/brave/vlc etc etc.

[V1tol]

Laptop + Nvidia Blob (RTX 2060) + Manjaro user here. PRIME is working for me both offloading to Nvidia as "external" GPU and running with Reverse PRIME.

There are problems though. For example - it stutters sometimes on desktop in "nvidia" mode (maybe some desync issue). Also I did not figure out how to run internal display on Intel and external displays on Nvidia simultaneously (external display ports are connected to Nvidia on my notebook, thus using "nvidia" mode). Another one known issue is Vulkan can hang for 10 seconds in "nvidia" mode if alt-tabbing for example from Proton game (most irritating issue, fixed in previous driver beta series, waiting in stable).

As for power consumption I use my notebook mostly as a desktop, so "nvidia" mode enabled. But my RTX 2060 is capable of new power management features (that one behind NVreg_DynamicPowerManagement flag) and "nvidia-smi" reports 1W consumption in "hybrid" mode (in "nvidia" mode idles on 8W). On battery I set up "intel" mode in optimus-manager (but never used it yet).